CVG-IBA among top ranked methods at MediaEval 2025

October 25-26, 2025: The School of Mathematics and Computer Science (SMCS), IBA Karachi is pleased to announce the outstanding performance of the Computer Vision Research Group at IBA (CVG-IBA) at MediaEval 2025, an esteemed international workshop focused on benchmarking and advancing computer vision research.

The final-year Computer Science students, along with an MSCS student, participated under the supervision of CS faculty members Dr. Muhammad Atif Tahir and Dr. Rizwan Ahmed Khan, securing top positions among leading international competitors.

Task 1: Movie Memorability | 1st Place in Prediction Challenge & Brand Memorability Challenge

Methodology: Exploring Visual, Textual, and Engagement Features for Memorability Predictions: Working Notes
Authors: Mahnoor Adeel, Kisa Fatima, Muhammad Ibrahim Ayoubi, Mustafa Usmani, and Dr. Muhammad Atif Tahir

By usage of the combination of all the specified methods, we, as a team, were able to not only predict very accurately which movies would be memorable but also which brands would be attached to the movies most strongly and in the end, we were able to be awarded diplomas for the first places at the international competition.

Task 2: Visual Question Answering for Gastrointestinal Imaging | 2nd Place in VQA Performance & Explainability Challenge

Methodology: Multi-Task Learning for Visually Grounded Reasoning in Gastrointestinal VQA
Authors: Itbaan Safwan, Annas Shaikh, Muhammad Haaris, Ramail Khan, and Dr. Muhammad Atif Tahir

The project proposed a multi-task learning framework that not only improved but also facilitated the clinical image interpretation with multimodal explanations as a novel contribution.

Task 3: News Images: Retrieval & Generative AI for News Thumbnails | 1st Place in Image Retrieval Challenge & Image Generation Challenge

Methodology: Balancing Relevance and Compliance: Text-to-Image Methods for Responsible Journalism
Authors: Mahrukh Khan, Alishba Subhani, Muhammad Rafi, and Dr. Muhammad Atif Tahir (In collaboration with FAST-NUCES, Karachi Campus)

The team designed responsible text-to-image generation pipelines that prioritize factual accuracy, ethical constraints, and semantic alignment—setting new benchmarks for AI-assisted journalism.

Task 4: Synthetic Images: Detecting Generative AI in Online Media | Top 5 in Classification Challenge

Methodology: Real versus Synthetic Classification using ResNet
Authors: Hafiz Muhammad Owais Raza, Dr. Muhammad Atif Tahir, and Dr. Rizwan Ahmed Khan

This work focused on improving real-versus-AI-generated image detection, contributing to global efforts in combating digital misinformation and synthetic media misuse.

These achievements reflect the devotion of SMCS and CVG-IBA to advancing computer vision research, and empowering students to compete and excel on international platforms.